New stack
intro
- As you may or may not know, I (hidnplayr) am since long time working on a new network stack for KolibriOS
- I will use this page for myself as a total image of what I have done so far, but also to inform others.
I'm currently working on:
* Complete rewrite of TCP code based on that of 4.4BSD
* Fragmenting IPv4 packets * updating netstat and netcfg applications * Learning more about MII and transceiver chips * Completing the dec21x4x driver * Completing/testing SLIP compatible RS232 driver * Finishing TFTP client
Kernel
The old kernel functions 52 and 53 have been removed and the New_network_api has been created.
Internally, the stack has been completely rewritten (mostly from scratch)
- What works
- Attaching network drivers to the kernel
- Sending and receiving Ethernet/IPv4/UDP/ICMP packets
- IPv4/UDP/TCP checksum generation and confirmation
- Queuing of packets (altough queing ethernet packets is disabled ATM)
- UDP sockets
- ~Working with multiple network card's (unfortunately, some kernel functions are hardcoded to a specific device ATM)
- TODO (items with a * are currently being worked on)
- Sending fragmented IPv4 packets
- finish TCP code*
- finish ARP code
- IPC sockets (unix-like sockets)
- RAW sockets
- Port trunking (IEEE 802.1AX-2008)
- ..
Drivers
These new type network drivers are MS COFF files, just like the sound drivers in KolibriOS.
- Notice that one driver can handle multiple devices.
Once the driver is loaded, an application (NetCFG) can use system function 68/17 ("Driver Control") to communicate with the driver.
For this, it uses the IOCTL structure wich looks like this:
<asm>
struc IOCTL {
.handle dd ? .io_code dd ? .input dd ? .inp_size dd ? .output dd ? .out_size dd ?
} </asm> for more info about this, please read Writing_drivers_for_KolibriOS
Here is a list of opcodes:
- 0 - Getversion
standard for all drivers, not only network drivers
- 1 - Hook
Attach driver to a device
IN: (ISA) <asm> type db 0 ; isa io_addr dw ? irq_line db ? </asm>
IN: (PCI) <asm> type db 1 ; pci pci_bus db ? pci_dev db ? </asm>
OUT:
eax = device number or -1 on error
Once the HOOK function of the driver has been called, the driver creates the NET_DEVICE structure, allocates some buffers needed for the driver, and calls the kernel function 'NetRegDev'.
This function registers a networkdevice (device, not driver!) to the kernel.
It only needs one parameter: pointer to device structure (NET_DEVICE) in ebx.
The function returns the device number in edi (..wich the driver then sends to the application in eax, see above..)
This NET_DEVICE structure is created and filled in by the driver, for every device it is hooked to.
The driver may use this to store pointers to buffers, descriptors,.. etc
But the top of the structure must always look like this: (Since it is shared by the driver and the kernel)
<asm> NET_DEVICE:
.type dd ? ; Type field (ethernet/slip/...)
.mtu dd ? ; Maximal Transmission Unit (in bytes)
.name dd ? ; Ptr to 0 terminated string
.unload dd ? ; Ptrs to driver functions
.reset dd ? ;
.transmit dd ? ;
.bytes_tx dq ? ; Statistics, updated by the driver
.bytes_rx dq ? ;
.packets_tx dd ? ;
.packets_rx dd ? ;
.end:
</asm>
This is appended by a Type-specific structure, for ethernet this will be:
<asm>
.set_mode dd ?
.get_mode dd ?
.set_MAC dd ?
.get_MAC dd ?
.mode dd ?
.mac dp ?
</asm>
Wich is then followed by the private data for the driver.
When a packet is received by the driver, it copies it into a buffer allocated using 'KernelAlloc'.
Then the driver pushes the size of the buffer and the pointer to the buffer onto the stack (in this particular order)
And it calls 'EthReceiver' (or another procedure for another network device type).
This kernel function will but the packet into the incoming queue, so the appropriate protocol handler can handle it later.
When the kernel needs to send a packet, it simple looks up the appropriate driver in it's driver table.
In the table it will find the pointer to the device structure, so the driver can directly call the transmit function listed in the structure.
NOTE: the kernel will put the address of buffer to send in [esp+4], and length of that buffer in [esp+8]
Current status of the drivers
- RTL8139: Working (FIXME: error occurs on wrapping of the RX buffer).
- SIS900: Working? (Written by clevermouse)
- 3c90x/3c59x: Working (only boomerang for now) (FIXME: may hang on UpUnstall procedure)
- RTL8029: Working
- PCnet32: Working
- DEC21x4x: Working in Virtual PC, but not on real hardware.
- i8255x (Intel eepro 100): Work in progress (hidnplayr)
- i8254x (Intel eepro 1000): Planned for future (hidnplayr)
- mtd80x: Planned for future (hidnplayr)
Programs
- NetCFG
- This is a program I wrote to load the drivers, it detects all network cards in your computer (PCI), and lets you load the appropriate driver for it.
You can run it with paramters 'F' to silently load first detected card, or parameter 'A' to load all cards
- Netstat
- This program allows you to read some variables from the stack.
- ARPcfg
- This program allows you to view the ARP entries, in future, you should be able to add and remove static entry's using this progam.
- Zeroconf
- This program is the newer version of what used to be 'autodhcp', it works with the new network API.
- nslookup
- Commandline DNS client from CleverMouse
uses network.obj
- TFTPc
- A TFTP client with a GUI.
can receive and send files, using libio.obj, network.obj and box_lib.obj
- Telnet
- a new telnet client, based on the code of nslookup (usefull to test TCP code)
uses network.obj
Upcoming
- ICMP
- program for sending ICMP echo reply's (ping) and trace-route.
Libraries
- network_lib
- The network library for KolibriOS, written by Clevermouse, was ported to the net branch, by the same person
I need your help!
I would really appreciate it if somebody could write some network applications,
so I can further test the kernel code without having to code them myself now :),
If you are interessed please contact me at hidnplayr@kolibrios.org (or find me in #general on irc.kolibrios.org)